Semi-Automatic Segmentation, Detection and Classification of Gram Stained Bacteria in Blood Samples

Abstract: Manual microscopy is a time-consuming and inefficient procedure in microbiology laboratories today. Common analyses in these laboratories are detection and classification of Gram stained bacteria (Rydberg J. personal communication. Feb. 2016). Bacteria that have been Gram stained are either Gram negative or Gram positive. Gram negative bacteria are pink/red and Gram positive bacteria are purple. Moreover, the bacteria can have different morphology. The most common are cocci and bacilli, cocci are round and bacilli are rod-shaped bacteria. Lastly, they are classified based on how they grow which, for instance, can be in chains or clusters. This thesis investigates whether it is possible to make an automatic, digital system that can replace manual microscopy for Gram stained bacteria. Images of bacteria were acquired with a digital microscope, provided by the company where the thesis was written, CellaVision AB. A method that segmented bacteria from background in the images was developed. Moreover, several methods have been implemented aiming to detect and classify bacteria based on their color, shape and arrangement. A final system was created that combined the most successful methods that enabled detection and classification of Gram stained bacteria. It could be concluded that an automatic, digital system for detection and classification of Gram stained bacteria is possible to implement. The system developed in this thesis was however semi-automatic since some user input was needed.